Edge - High Scalability -

Olden Days

10 or so years ago
Mostly 1 gbps network.
State of the art was a monolithic Java application connected to a big relational database.
Systems were decomposed using Web Services that exchange XML messages.
Sending XML over a slow network had high overhead. Just parsing XML is slow.
Since web services were slow the resulting architecture featured large servers with a relatively low number of messages exchanged between them.
The slowness of the system meant a request couldn’t hit more than two or three services before returning a reply, so it was difficult to decompose a monolith into smaller services.

Now

Network is 25 gbps.
Messages are Avro, gRPC, etc, which are 100x more efficient than XML.
Combining the faster network with better message formats means we’re 100x to 1000x more efficient at sending messages between servers. (Historical note: we’ve always had cheap messaging on Unix, people just forgot that when they went insane with XML.)
Now it’s possible to break an architecture up into many smaller services because it’s possible to make 100 microservices calls and return a reply within latency bounds.
Microservices became possible because of a technology change. A system could be broken up into smaller services where every service has a single responsibility.
Now it’s possible to a go a step further by breaking up microservices into the separate functions making up those services.
We can afford to connect all these functions together because messaging has become cheaper.